search

rails / solid_queue

Database-backed Active Job backend
MIT License
1.95k stars 130 forks source link

readme

Solid Queue

Solid Queue is a DB-based queuing backend for Active Job, designed with simplicity and performance in mind.

Besides regular job enqueuing and processing, Solid Queue supports delayed jobs, concurrency controls, recurring jobs, pausing queues, numeric priorities per job, priorities by queue order, and bulk enqueuing (enqueue_all for Active Job's perform_all_later).

Solid Queue can be used with SQL databases such as MySQL, PostgreSQL or SQLite, and it leverages the FOR UPDATE SKIP LOCKED clause, if available, to avoid blocking and waiting on locks when polling jobs. It relies on Active Job for retries, discarding, error handling, serialization, or delays, and it's compatible with Ruby on Rails's multi-threading.

Installation

Solid Queue is configured by default in new Rails 8 applications. But if you're running an earlier version, you can add it manually following these steps:

  1. bundle add solid_queue
  2. bin/rails solid_queue:install

This will configure Solid Queue as the production Active Job backend, create the configuration files config/queue.yml and config/recurring.yml, and create the db/queue_schema.rb. It'll also create a bin/jobs executable wrapper that you can use to start Solid Queue.

Once you've done that, you will then have to add the configuration for the queue database in config/database.yml. If you're using SQLite, it'll look like this:

production:
  primary:
    <<: *default
    database: storage/production.sqlite3
  queue:
    <<: *default
    database: storage/production_queue.sqlite3
    migrations_paths: db/queue_migrate

...or if you're using MySQL/PostgreSQL/Trilogy:

production:
  primary: &primary_production
    <<: *default
    database: app_production
    username: app
    password: <%= ENV["APP_DATABASE_PASSWORD"] %>
  queue:
    <<: *primary_production
    database: app_production_queue
    migrations_paths: db/queue_migrate

Note: Calling bin/rails solid_queue:install will automatically add config.solid_queue.connects_to = { database: { writing: :queue } } to config/environments/production.rb, so no additional configuration is needed there (although you must make sure that you use the queue name in database.yml for this to match!). But if you want to use Solid Queue in a different environment (like staging or even development), you'll have to manually add that config.solid_queue.connects_to line to the respective environment file. And, as always, make sure that the name you're using for the database in config/database.yml matches the name you use in config.solid_queue.connects_to.

Then run db:prepare in production to ensure the database is created and the schema is loaded.

Now you're ready to start processing jobs by running bin/jobs on the server that's doing the work. This will start processing jobs in all queues using the default configuration. See below to learn more about configuring Solid Queue.

For small projects, you can run Solid Queue on the same machine as your webserver. When you're ready to scale, Solid Queue supports horizontal scaling out-of-the-box. You can run Solid Queue on a separate server from your webserver, or even run bin/jobs on multiple machines at the same time. Depending on the configuration, you can designate some machines to run only dispatchers or only workers. See the configuration section for more details on this.

Note: future changes to the schema will come in the form of regular migrations.

Single database configuration

Running Solid Queue in a separate database is recommended, but it's also possible to use one single database for both the app and the queue. Just follow these steps:

  1. Copy the contents of db/queue_schema.rb into a normal migration and delete db/queue_schema.rb
  2. Remove config.solid_queue.connects_to from production.rb
  3. Migrate your database. You are ready to run bin/jobs

You won't have multiple databases, so database.yml doesn't need to have primary and queue database.

Incremental adoption

If you're planning to adopt Solid Queue incrementally by switching one job at the time, you can do so by leaving the config.active_job.queue_adapter set to your old backend, and then set the queue_adapter directly in the jobs you're moving:

# app/jobs/my_job.rb

class MyJob < ApplicationJob
  self.queue_adapter = :solid_queue
  # ...
end

High performance requirements

Solid Queue was designed for the highest throughput when used with MySQL 8+ or PostgreSQL 9.5+, as they support FOR UPDATE SKIP LOCKED. You can use it with older versions, but in that case, you might run into lock waits if you run multiple workers for the same queue. You can also use it with SQLite on smaller applications.

Configuration

Workers, dispatchers and scheduler

We have several types of actors in Solid Queue:

Solid Queue's supervisor will fork a separate process for each supervised worker/dispatcher/scheduler.

By default, Solid Queue will try to find your configuration under config/queue.yml, but you can set a different path using the environment variable SOLID_QUEUE_CONFIG or by using the -c/--config_file option with bin/jobs, like this:

bin/jobs -c config/calendar.yml

This is what this configuration looks like:

production:
  dispatchers:
    - polling_interval: 1
      batch_size: 500
      concurrency_maintenance_interval: 300
  workers:
    - queues: "*"
      threads: 3
      polling_interval: 2
    - queues: [ real_time, background ]
      threads: 5
      polling_interval: 0.1
      processes: 3

Everything is optional. If no configuration at all is provided, Solid Queue will run with one dispatcher and one worker with default settings. If you want to run only dispatchers or workers, you just need to include that section alone in the configuration. For example, with the following configuration:

production:
  dispatchers:
    - polling_interval: 1
      batch_size: 500
      concurrency_maintenance_interval: 300

the supervisor will run 1 dispatcher and no workers.

Here's an overview of the different options:

Queue order and priorities

As mentioned above, if you specify a list of queues for a worker, these will be polled in the order given, such as for the list real_time,background, no jobs will be taken from background unless there aren't any more jobs waiting in real_time.

Active Job also supports positive integer priorities when enqueuing jobs. In Solid Queue, the smaller the value, the higher the priority. The default is 0.

This is useful when you run jobs with different importance or urgency in the same queue. Within the same queue, jobs will be picked in order of priority, but in a list of queues, the queue order takes precedence, so in the previous example with real_time,background, jobs in the real_time queue will be picked before jobs in the background queue, even if those in the background queue have a higher priority (smaller value) set.

We recommend not mixing queue order with priorities but either choosing one or the other, as that will make job execution order more straightforward for you.

Queues specification and performance

To keep polling performant and ensure a covering index is always used, Solid Queue only does two types of polling queries:

-- No filtering by queue
SELECT job_id
FROM solid_queue_ready_executions
ORDER BY priority ASC, job_id ASC
LIMIT ?
FOR UPDATE SKIP LOCKED;

-- Filtering by a single queue
SELECT job_id
FROM solid_queue_ready_executions
WHERE queue_name = ?
ORDER BY priority ASC, job_id ASC
LIMIT ?
FOR UPDATE SKIP LOCKED;

The first one (no filtering by queue) is used when you specify

queues: *

and there aren't any queues paused, as we want to target all queues.

In other cases, we need to have a list of queues to filter by, in order, because we can only filter by a single queue at a time to ensure we use an index to sort. This means that if you specify your queues as:

queues: beta*

we'll need to get a list of all existing queues matching that prefix first, with a query that would look like this:

SELECT DISTINCT(queue_name)
FROM solid_queue_ready_executions
WHERE queue_name LIKE 'beta%';

This type of DISTINCT query on a column that's the leftmost column in an index can be performed very fast in MySQL thanks to a technique called Loose Index Scan. PostgreSQL and SQLite, however, don't implement this technique, which means that if your solid_queue_ready_executions table is very big because your queues get very deep, this query will get slow. Normally your solid_queue_ready_executions table will be small, but it can happen.

Similarly to using prefixes, the same will happen if you have paused queues, because we need to get a list of all queues with a query like

SELECT DISTINCT(queue_name)
FROM solid_queue_ready_executions

and then remove the paused ones. Pausing in general should be something rare, used in special circumstances, and for a short period of time. If you don't want to process jobs from a queue anymore, the best way to do that is to remove it from your list of queues.

đź’ˇ To sum up, if you want to ensure optimal performance on polling, the best way to do that is to always specify exact names for them, and not have any queues paused.

Do this:

queues: background, backend

instead of this:

queues: back*

Threads, processes and signals

Workers in Solid Queue use a thread pool to run work in multiple threads, configurable via the threads parameter above. Besides this, parallelism can be achieved via multiple processes on one machine (configurable via different workers or the processes parameter above) or by horizontal scaling.

The supervisor is in charge of managing these processes, and it responds to the following signals:

When receiving a QUIT signal, if workers still have jobs in-flight, these will be returned to the queue when the processes are deregistered.

If processes have no chance of cleaning up before exiting (e.g. if someone pulls a cable somewhere), in-flight jobs might remain claimed by the processes executing them. Processes send heartbeats, and the supervisor checks and prunes processes with expired heartbeats, which will release any claimed jobs back to their queues. You can configure both the frequency of heartbeats and the threshold to consider a process dead. See the section below for this.

Database configuration

You can configure the database used by Solid Queue via the config.solid_queue.connects_to option in the config/application.rb or config/environments/production.rb config files. By default, a single database is used for both writing and reading called queue to match the database configuration you set up during the install.

All the options available to Active Record for multiple databases can be used here.

Lifecycle hooks

In Solid queue, you can hook into two different points in the supervisor's life:

And into two different points in a worker's life:

You can use the following methods with a block to do this:

SolidQueue.on_start
SolidQueue.on_stop

SolidQueue.on_worker_start
SolidQueue.on_worker_stop

For example:

SolidQueue.on_start { start_metrics_server }
SolidQueue.on_stop { stop_metrics_server }

These can be called several times to add multiple hooks, but it needs to happen before Solid Queue is started. An initializer would be a good place to do this.

Other configuration settings

Note: The settings in this section should be set in your config/application.rb or your environment config like this: config.solid_queue.silence_polling = true

There are several settings that control how Solid Queue works that you can set as well:

Errors when enqueuing

Solid Queue will raise a SolidQueue::Job::EnqueueError for any Active Record errors that happen when enqueuing a job. The reason for not raising ActiveJob::EnqueueError is that this one gets handled by Active Job, causing perform_later to return false and set job.enqueue_error, yielding the job to a block that you need to pass to perform_later. This works very well for your own jobs, but makes failure very hard to handle for jobs enqueued by Rails or other gems, such as Turbo::Streams::BroadcastJob or ActiveStorage::AnalyzeJob, because you don't control the call to perform_later in that cases.

In the case of recurring tasks, if such error is raised when enqueuing the job corresponding to the task, it'll be handled and logged but it won't bubble up.

Concurrency controls

Solid Queue extends Active Job with concurrency controls, that allows you to limit how many jobs of a certain type or with certain arguments can run at the same time. When limited in this way, jobs will be blocked from running, and they'll stay blocked until another job finishes and unblocks them, or after the set expiry time (concurrency limit's duration) elapses. Jobs are never discarded or lost, only blocked.

class MyJob < ApplicationJob
  limits_concurrency to: max_concurrent_executions, key: ->(arg1, arg2, **) { ... }, duration: max_interval_to_guarantee_concurrency_limit, group: concurrency_group

  # ...

When a job includes these controls, we'll ensure that, at most, the number of jobs (indicated as to) that yield the same key will be performed concurrently, and this guarantee will last for duration for each job enqueued. Note that there's no guarantee about the order of execution, only about jobs being performed at the same time (overlapping).

The concurrency limits use the concept of semaphores when enqueuing, and work as follows: when a job is enqueued, we check if it specifies concurrency controls. If it does, we check the semaphore for the computed concurrency key. If the semaphore is open, we claim it and we set the job as ready. Ready means it can be picked up by workers for execution. When the job finishes executing (be it successfully or unsuccessfully, resulting in a failed execution), we signal the semaphore and try to unblock the next job with the same key, if any. Unblocking the next job doesn't mean running that job right away, but moving it from blocked to ready. Since something can happen that prevents the first job from releasing the semaphore and unblocking the next job (for example, someone pulling a plug in the machine where the worker is running), we have the duration as a failsafe. Jobs that have been blocked for more than duration are candidates to be released, but only as many of them as the concurrency rules allow, as each one would need to go through the semaphore dance check. This means that the duration is not really about the job that's enqueued or being run, it's about the jobs that are blocked waiting.

For example:

class DeliverAnnouncementToContactJob < ApplicationJob
  limits_concurrency to: 2, key: ->(contact) { contact.account }, duration: 5.minutes

  def perform(contact)
    # ...

Where contact and account are ActiveRecord records. In this case, we'll ensure that at most two jobs of the kind DeliverAnnouncementToContact for the same account will run concurrently. If, for any reason, one of those jobs takes longer than 5 minutes or doesn't release its concurrency lock (signals the semaphore) within 5 minutes of acquiring it, a new job with the same key might gain the lock.

Let's see another example using group:

class Box::MovePostingsByContactToDesignatedBoxJob < ApplicationJob
  limits_concurrency key: ->(contact) { contact }, duration: 15.minutes, group: "ContactActions"

  def perform(contact)
    # ...
class Bundle::RebundlePostingsJob < ApplicationJob
  limits_concurrency key: ->(bundle) { bundle.contact }, duration: 15.minutes, group: "ContactActions"

  def perform(bundle)
    # ...

In this case, if we have a Box::MovePostingsByContactToDesignatedBoxJob job enqueued for a contact record with id 123 and another Bundle::RebundlePostingsJob job enqueued simultaneously for a bundle record that references contact 123, only one of them will be allowed to proceed. The other one will stay blocked until the first one finishes (or 15 minutes pass, whatever happens first).

Note that the duration setting depends indirectly on the value for concurrency_maintenance_interval that you set for your dispatcher(s), as that'd be the frequency with which blocked jobs are checked and unblocked. In general, you should set duration in a way that all your jobs would finish well under that duration and think of the concurrency maintenance task as a failsafe in case something goes wrong.

Jobs are unblocked in order of priority but queue order is not taken into account for unblocking jobs. That means that if you have a group of jobs that share a concurrency group but are in different queues, or jobs of the same class that you enqueue in different queues, the queue order you set for a worker is not taken into account when unblocking blocked ones. The reason is that a job that runs unblocks the next one, and the job itself doesn't know about a particular worker's queue order (you could even have different workers with different queue orders), it can only know about priority. Once blocked jobs are unblocked and available for polling, they'll be picked up by a worker following its queue order.

Finally, failed jobs that are automatically or manually retried work in the same way as new jobs that get enqueued: they get in the queue for getting an open semaphore, and whenever they get it, they'll be run. It doesn't matter if they had already gotten an open semaphore in the past.

Failed jobs and retries

Solid Queue doesn't include any automatic retry mechanism, it relies on Active Job for this. Jobs that fail will be kept in the system, and a failed execution (a record in the solid_queue_failed_executions table) will be created for these. The job will stay there until manually discarded or re-enqueued. You can do this in a console as:

failed_execution = SolidQueue::FailedExecution.find(...) # Find the failed execution related to your job
failed_execution.error # inspect the error

failed_execution.retry # This will re-enqueue the job as if it was enqueued for the first time
failed_execution.discard # This will delete the job from the system

However, we recommend taking a look at mission_control-jobs, a dashboard where, among other things, you can examine and retry/discard failed jobs.

Error reporting on jobs

Some error tracking services that integrate with Rails, such as Sentry or Rollbar, hook into Active Job and automatically report not handled errors that happen during job execution. However, if your error tracking system doesn't, or if you need some custom reporting, you can hook into Active Job yourself. A possible way of doing this would be:

# application_job.rb
class ApplicationJob < ActiveJob::Base
  rescue_from(Exception) do |exception|
    Rails.error.report(exception)
    raise exception
  end
end

Note that, you will have to duplicate the above logic on ActionMailer::MailDeliveryJob too. That is because ActionMailer doesn't inherit from ApplicationJob but instead uses ActionMailer::MailDeliveryJob which inherits from ActiveJob::Base.

# application_mailer.rb

class ApplicationMailer < ActionMailer::Base
  ActionMailer::MailDeliveryJob.rescue_from(Exception) do |exception|
    Rails.error.report(exception)
    raise exception
  end
end

Puma plugin

We provide a Puma plugin if you want to run the Solid Queue's supervisor together with Puma and have Puma monitor and manage it. You just need to add

plugin :solid_queue

to your puma.rb configuration.

Jobs and transactional integrity

:warning: Having your jobs in the same ACID-compliant database as your application data enables a powerful yet sharp tool: taking advantage of transactional integrity to ensure some action in your app is not committed unless your job is also committed and vice versa, and ensuring that your job won't be enqueued until the transaction within which you're enqueuing it is committed. This can be very powerful and useful, but it can also backfire if you base some of your logic on this behaviour, and in the future, you move to another active job backend, or if you simply move Solid Queue to its own database, and suddenly the behaviour changes under you. Because this can be quite tricky and many people shouldn't need to worry about it, by default Solid Queue is configured in a different database as the main app.

Starting from Rails 8, an option which doesn't rely on this transactional integrity and which Active Job provides is to defer the enqueueing of a job inside an Active Record transaction until that transaction successfully commits. This option can be set via the enqueue_after_transaction_commit class method on the job level and is by default disabled. Either it can be enabled for individual jobs or for all jobs through ApplicationJob:

class ApplicationJob < ActiveJob::Base
  self.enqueue_after_transaction_commit = true
end

Using this option, you can also use Solid Queue in the same database as your app but not rely on transactional integrity.

If you don't set this option but still want to make sure you're not inadvertently on transactional integrity, you can make sure that:

Recurring tasks

Solid Queue supports defining recurring tasks that run at specific times in the future, on a regular basis like cron jobs. These are managed by the scheduler process and are defined in their own configuration file. By default, the file is located in config/recurring.yml, but you can set a different path using the environment variable SOLID_QUEUE_RECURRING_SCHEDULE or by using the --recurring_schedule_file option with bin/jobs, like this:

bin/jobs --recurring_schedule_file=config/schedule.yml

The configuration itself looks like this:

production:
  a_periodic_job:
    class: MyJob
    args: [ 42, { status: "custom_status" } ]
    schedule: every second
  a_cleanup_task:
    command: "DeletedStuff.clear_all"
    schedule: every day at 9am

Tasks are specified as a hash/dictionary, where the key will be the task's key internally. Each task needs to either have a class, which will be the job class to enqueue, or a command, which will be eval'ed in the context of a job (SolidQueue::RecurringJob) that will be enqueued according to its schedule, in the solid_queue_recurring queue.

Each task needs to have also a schedule, which is parsed using Fugit, so it accepts anything that Fugit accepts as a cron. You can optionally supply the following for each task:

The job in the example configuration above will be enqueued every second as:

MyJob.perform_later(42, status: "custom_status")

Tasks are enqueued at their corresponding times by the scheduler, and each task schedules the next one. This is pretty much inspired by what GoodJob does.

It's possible to run multiple schedulers with the same recurring_tasks configuration, for example, if you have multiple servers for redundancy, and you run the scheduler in more than one of them. To avoid enqueuing duplicate tasks at the same time, an entry in a new solid_queue_recurring_executions table is created in the same transaction as the job is enqueued. This table has a unique index on task_key and run_at, ensuring only one entry per task per time will be created. This only works if you have preserve_finished_jobs set to true (the default), and the guarantee applies as long as you keep the jobs around.

Note: a single recurring schedule is supported, so you can have multiple schedulers using the same schedule, but not multiple schedulers using different configurations.

Finally, it's possible to configure jobs that aren't handled by Solid Queue. That is, you can have a job like this in your app:

class MyResqueJob < ApplicationJob
  self.queue_adapter = :resque

  def perform(arg)
    # ..
  end
end

You can still configure this in Solid Queue:

my_periodic_resque_job:
  class: MyResqueJob
  args: 22
  schedule: "*/5 * * * *"

and the job will be enqueued via perform_later so it'll run in Resque. However, in this case we won't track any solid_queue_recurring_execution record for it and there won't be any guarantees that the job is enqueued only once each time.

Inspiration

Solid Queue has been inspired by resque and GoodJob. We recommend checking out these projects as they're great examples from which we've learnt a lot.

License

The gem is available as open source under the terms of the MIT License.